I used pygame for a reinforcement learning course assignment: I built a gridworld simulator where you can input a text file with the origin, goal and obstacles, and watch as an agent learns a policy using Q-learning. It's based on a visualization used in a Berkeley RL bootcamp: https://miro.medium.com/max/992/1*FnCKMEFZC9QRnxLWT3lSCA.png

I never got to open-sourcing it though